Use of carbon dioxide to prevent zebra mussel (Dreissena polymorpha) settlement and effects on native mussels (Order Unionoida) and benthic communities

We determined the efficacy of carbon dioxide (CO2) for preventing larval (veliger) settlement of the invasive zebra mussel (Dreissena polymorpha) and compared the response of native juvenile mussels and macroinvertebrate and periphyton communities to hypercapnia. A microcosm system of control (ambient Camb) and CO2 treatments [low partial pressure of carbon dioxide, PCO2 (Clow) = 35,094 μatm, and high PCO2 (Chigh) = 66,685 μatm)] was supplied with river water for 11 weeks. We measured zebra mussel settlement, juvenile native mussel (Lampsilis cardium and L. siliquoidea) growth and condition, and macroinvertebrate and periphyton community composition. Zebra mussels settled in all Camb tanks (n = 4) from early July through August compared to one mussel in a CO2 treatment tank over the same period. Native mussel growth, tissue condition, and shell condition were reduced in CO2, particularly in the highest treatment. Benthic biomass (excluding zebra mussels) was similar in Camb and Clow but was reduced in Chigh. Macroinvertebrate community composition differed among treatments due to greater abundance of Chironomidae in Clow and Oligochaeata in Chigh. Periphyton abundance and richness increased in both CO2 treatments and was driven by increases in Cyanobacteria, Bacillariophyta, and Chlorophyta. Our results indicate that efficacious levels of CO2 (~ 35,000 μatm PCO2) for reducing biofouling by dreissenids are tolerable to most freshwater benthic taxa. Chronic elevated CO2 infusion (> 35,000 μatm PCO2) may reduce native mussel growth and condition and alter benthic invertebrate and periphyton community composition. Further refinement of CO2 application strategies is needed to determine minimum effective dose and duration to prevent dreissenid settlement and minimize effects on nontarget organisms.


Introduction
Zebra mussels, Dreissena polymorpha (Pallas, 1771) and quagga mussels, D. bugensis, (Andrusov, 1897) are invasive bivalves to North America that were introduced in the 1980s presumably in ballast water of intercontinental ships entering the Great Lakes (Griffiths et al. 1991). Unlike native unionid mussels, dreissenids are epifaunal and secrete adhesive byssal threads for attachment to substrates. Their prolific Citation: Waller DL, Bartsch L, Bartsch MR, Meulemans M, Severson T, Zolper TJ (2021) Use of carbon dioxide to prevent zebra mussel (Dreissena polymorpha) settlement and effects on native mussels (Order Unionoida) and benthic communities. Management of Biological Invasions 12(4): 927-951, https://doi.org/10. 3391/mbi.2021.12.4.10 reproduction, microscopic larvae (herein referred to as veligers), and rapid growth rate enabled dreissenid mussels to quickly spread and establish in waterbodies across North America (Mackie 1991;Benson et al. 2020). As highly efficient filter feeders, dreissenids have substantially altered trophic structures and nutrient cycles in freshwater systems (Strayer et al. 1998;Madenjian et al. 2015) and caused the decline and extirpation of some native species, notably unionid mussels (Vanderploeg et al. 2002;Higgins and Van der Zanden 2010;Bootsma and Liao 2014;Mayer et al. 2014). Users of freshwater for industry, agriculture, municipal drinking water, and hydroelectric generation are especially affected by dreissenid fouling. Mussels cause reduced water flow through water intake and cooling systems, deterioration of surfaces, and contamination of potable water when mussels die. As evidence, water managers in North America have spent an estimated $100,000 to millions of dollars per year to mitigate biofouling by dreissenids (Rosaen et al. 2016;Chakraborti et al. 2016).
Prevention and mitigation of macrofouling by mussels have focused primarily on chemical tools, especially oxidizing chemicals (e.g., chlorine), potassium compounds (permanganate, potassium hydroxide, potassium chloride; Mackie and Claudi 2010;Chakraborti et al. 2014;Davis et al. 2018), copper-based compounds (Claudi et al. 2014;Watters et al. 2013), and the biopesticide Zequanox® (Whitledge et al. 2015;Rackl and Link 2015). These tools are efficacious, but not without ecological and/or economic costs. Chlorine is commonly used in closed water systems for dreissenid control and has advantages of proven efficacy, cost-effectiveness, and simple application systems. However, chlorine can produce toxic byproducts, such as trihalomethane, poses a human safety concern in some forms, and must be neutralized before discharge into the environment (Mackie and Claudi 2010). Potassium and copper-based compounds have been used in both closed systems (Watters et al. 2013;Chakraborti et al. 2014) and open water (Offutt Air Force Base 2009;DFO 2014;Fernald and Watson 2014;Lund et al. 2017;Hammond and Ferris 2019). When used in open water, potassium and copper-based compounds can persist in the environment and may be toxic to other aquatic life at levels used in dreissenid control treatments (Offutt Air Force Base 2009;Fieldseth and Sweet 2016;Densmore et al. 2018).
Carbon dioxide (CO 2 ) was first tested in the 1990s as a molluscicide and found to effectively reduce attachment and kill zebra mussels (Elzinga and Butzlaff 1994;McMahon et al. 1995;Payne et al. 1998). More recently, CO 2 treatment regimens were determined for detachment and lethality of adult zebra mussels at different water temperatures (Waller and Bartsch 2018;Waller et al. 2020). The lethal time to 99% mortality (LT99) of adult zebra mussels in 110,000 to 120,000 µatm partial pressure of carbon dioxide (PCO 2 ) ranged from 100 h at 20 °C to 300 h at 5 °C, and short-term infusion (48 h) of CO 2 at this level reduced attachment by 50% at both temperatures . A next step in an evaluation of CO 2 for dreissenid control is to determine treatments that are efficacious for the more sensitive veliger stage. Metamorphosis and settlement of veligers to the juvenile stage is reportedly prevented in water with pH < 7.1 (Claudi et al. 2012). Carbon dioxide reacts in water to form carbonic acid, causing mild acidification. Therefore, we expect that CO 2 levels that decrease pH to < 7.1 will prevent veliger settlement and effectively prevent biofouling of water infrastructures.
Our study site was located at a harbor that is used by the U.S. Fish and Wildlife Service (FWS) Genoa National Fish Hatchery (GNFH) for seasonal propagation of federally-and state-listed endangered native mussel species. Propagation cages and juvenile native mussels become infested with zebra mussels which increases maintenance costs and reduces juvenile production.
As several studies have reported greater selective toxicity of CO 2 to zebra mussels compared to native mussels, we considered whether CO 2 could be used to prevent biofouling of native mussels by dreissenids. For example, Waller et al. (2020) found short-term PCO 2 treatments (maximum 300 h; 100,000 to 120,000 µatm) at 5, 12 and 20 °C that were lethal to adult zebra mussels and produced minimal mortality of juvenile Lampsilis cardium and Leptodea fragilis. However, other studies reported lethal and sublethal effects on juveniles of several species (Hannan et al. 2016a, b;Jeffrey et al. 2017Jeffrey et al. , 2018aWaller et al. 2017Waller et al. , 2019Waller and Bartsch 2018). Survival of Lampsilis spp. was significantly reduced after 14-d exposure to 58,000 µatm PCO 2 (Jeffrey et al. 2018b) or 28-d exposure to 30,000 to 58,000 µatm PCO 2 (Waller et al. 2019). Physiological indicators of stress and reduced shell growth and condition from elevated CO 2 have also been reported (Waller et al. 2017(Waller et al. , 2019Jeffrey et al. 2018b). More investigation is needed to establish safe levels of CO 2 for juvenile native mussels and to determine whether CO 2 could be a feasible option to reduce settlement of dreissenids in the vicinity of native mussels.
In addition to native mussels, other aquatic organisms in a natural water supply may be affected by elevated CO 2 . Several studies have evaluated the effects of elevated CO 2 on nontarget fish (Tix et al. 2017;Schneider et al. 2019;Cupp et al. 2020), but few have monitored the response of nontarget freshwater communities to CO 2 at levels required for AIS control actions (~ 50,000 to 100,000 µatm PCO 2 ) (Treanor et al. 2017;Hasler et al. 2018). Levels of PCO 2 that prevent dreissenid settlement may also reduce colonization of pH-sensitive native species, especially calcifying organisms, and shift macroinvertebrate and periphyton community composition to more tolerant taxa.
In the present study, we developed a flow-through microcosm test system to assess the effectiveness of CO 2 for preventing the settlement of zebra mussel veligers in a water intake or confined open water system (e.g., marina, culture pond). We used a raw river water supply that continually seeded the system with phytoplankton, various life stages of zooplankton and benthic invertebrates, and zebra mussel veligers. We assessed changes in nontarget benthic taxa to levels of CO 2 that were predicted to reduce zebra mussel settlement. Our objectives were to: (1) determine the effectiveness of CO 2 to prevent veliger settlement, (2) evaluate lethal and sublethal responses of native juvenile mussels to chronic CO 2 exposure, and (3) describe shifts in resident benthic macroinvertebrate and periphyton communities to chronic CO 2 exposure at levels used in AIS control actions.

Materials and methods
Our test system was housed within a mobile laboratory and withdrew water from a harbor of the Mississippi River, Dubuque, IA, to supply a natural source of veligers and plankton to colonize the tanks, similar to that described in Claudi et al. (2012). Water was pumped from the adjacent harbor into a headbox (132 cm, length × 79 cm, width × 36 cm, height). A gravity-flow headbox outflowed to three infusion chambers (101.6 mm, diameter × 38 cm, length; Supplementary material Figures S1 and S2) where gas was mixed and dissolved in the water. Two chambers were infused with CO 2 and one chamber was infused with atmospheric air. We created a calibration curve of pH versus CO 2 concentration in the source water and then regulated CO 2 flow into each chamber based on pH probe measurements in the infusion chambers ( Figure S2). Alkalinity was measured in the source water every week, and we recalibrated the pH-CO 2 curve when source water alkalinity changed by ± 50 mg/L. Target CO 2 concentrations for the low (C low ) and high (C high ) treatments were 30,000 µatm PCO 2 (~ 50 mg/L CO 2 ) and 55,000 µatm PCO 2 (~ 90 mg/L CO 2 ), respectively. The low CO 2 level corresponded to pH 6.9, just below the pH reported to prevent veliger settlement (Claudi et al. 2012). The high CO 2 was a maximum level expected to be used in long-term CO 2 AIS application for dreissenid control.
Each infusion chamber (ambient, low, and high CO 2 ) delivered water to four replicate tanks (66 cm, length × 30.5 cm, width × 30.5 cm, height; 51 L volume) per treatment (3 treatments × 4 replicates = 12 tank total). Flowrate to each tank produced about 1.5 tank exchanges per hour. Ancillary measurements (i.e., water quality, flowrate, temperature) were taken regularly to identify potential noncausal influences on tank position.

Settlement plate assessment
Four settlement samplers were placed into each test tank about 2 weeks before the onset of CO 2 infusion to establish invertebrate and periphyton communities on the plates. Sampler design was modified from Pucherelli and Claudi (2017) and consisted of six PVC plates (12 cm, length × 12 cm, width × 6 mm, thickness) mounted vertically into a slotted base (24 cm, length × 12 cm, width × 1.3 cm, thickness) and spaced ~ 3.0 cm apart ( Figure S2). We collected settlement plates from two replicate tanks per treatment on alternate weeks beginning at week 4 (July 3) and extending to week 11 (August 19). During each sampling period, we removed one plate from each of 4 samplers within a given tank, gently scraped both sides of the plate with a razor blade, and returned the plate to the tank to maintain the geometry of the sampler. Each plate was sampled only once during the experiment. We composited the scrapings from the four plates per tank making a single sample per tank. Samples were preserved in ethanol for enumeration of zebra mussels and macroinvertebrates. On the last sampling date, we scraped four plates from all tanks. Macroinvertebrate identification and biomass measurements were conducted by Rhithron Associates, Inc., Missoula, Montana, USA.
Preserved macroinvertebrates were counted and identified to genus level, except Chironomidae which were identified to family level. If more than 300 organisms were present in the sample, macroinvertebrates were subsampled using a grid tray. If less than 300 organisms were present, the entire sample was identified. The subsample that was removed for identification was returned to the composite sample for determination of dry weight. Zebra mussels were separated from all other organisms for enumeration and determination of dry weight. Samples were divided into two mass groups (zebra mussel and "non-zebra mussel"), dried at 60 °C to a constant mass, and dried biomass (0.1 g) was measured. In C low and C high tanks, total biomass was equal to non-zebra mussel biomass. In C amb tanks, we determined total biomass from the sum of zebra mussel and non-zebra mussel biomass.
Periphyton was collected on the last sampling day by scraping one plate per sampler in each test tank (n = 4 plates; n = 12 tanks). Scrapings from the four plates were combined, as described for benthic invertebrate sampling, and preserved with 2% Lugols' solution. Periphyton analysis was conducted by BSA Environmental Services, Inc., Beachwood, Ohio, USA. Briefly, cell numbers of all identified algal and cyanobacteria taxa were quantified and converted to cell density (cells per cm 2 ) and biovolume (µm³/L) using either the Utermohl method (Lund et al. 1958) or the membrane filtration technique (McNabb 1960). The abundance of common taxa was estimated by random field counts of 300 to 400 natural units (colonies, filaments, unicells) or a minimum of 50 random fields enumerated to the lowest possible taxonomic level.

Native mussels
Juvenile (1-yr-old) L. cardium (Rafinesque, 1820) were obtained from the FWS Genoa National Fish Hatchery, Genoa, Wisconsin, USA, and L. siliquoidea (Barnes, 1823) were obtained from the U.S. Geological Survey Upper Midwest Environmental Sciences Center, La Crosse, Wisconsin, USA. Before placement into tanks, we took digital images of individuals (0.75×) with a Nikon CoolPix500 camera mounted on a Nikon SMZ1500 microscope. Shell length, defined as the maximum anterior-posterior axis parallel to the hinge line, was measured on digital images using ImageJ software (nearest 0.02 mm). Mean (standard deviation, SD) initial shell length was 15.0 (2.0) and 10.2 (1.8) mm for L. cardium and L. siliquoidea, respectively. We placed 10 L. cardium and 20 L. siliquoidea into one container per tank (15 cm, diameter × 10 cm, depth) and provided ~ 7.5 cm depth of washed river sand for mussels to orient and bury ( Figure S2).
At the midpoint (day 40) and end (day 75) of the study, juvenile mussels were removed from containers and assessed for survival and growth. We measured shell length with a digital caliper (Mitutoyo CD-6" ASX, Mitutoyo manufacturing, Tokyo, Japan) (nearest 0.01 mm) because mussels were too large to photograph under the stereomicroscope. At the conclusion of the study, all L. siliquoidea (n = 20) and two L. cardium per tank were sacrificed for determination of dry weight. The remaining L. cardium (n = 8) were preserved for biochemical and transcriptomic analysis as part of a companion study. We separated soft tissue and shell, dried samples at 60 °C to a constant mass and measured dry mass (0.001 mg) on an analytical balance (AT200, Mettler, Columbus, Ohio, USA). We calculated condition indices to compare the effects of CO 2 on shell and soft tissue. Tissue condition (TC) was calculated as: TC = (dry tissue mass/shell length) × 100. Shell condition (SC) was calculated as: SC = (dry shell mass/shell length) × 100.

Carbon dioxide and water quality
We measured CO 2 concentration in test tanks by three methods: (1) the partial pressure of CO 2 (PCO 2 ) was calculated using tank pH and temperature, and headbox alkalinity values (Robbins et al. 2010), (2) %CO 2 was measured with a CO 2 probe (Vaisala BMP220 and GMT221, Vaisala Corp., Helsinki, Finland) and converted to PCO 2 based on temperature and barometric pressure, (3) CO 2 concentration was measured daily in each tank by titration with 3.636 N sodium hydroxide (NaOH) to a pH endpoint of 8.3 (APHA 2018a).
We measured water quality (dissolved oxygen, pH, temperature) and flow rate at least once a day in each test tank throughout the study. Dissolved oxygen (DO) and pH were measured with LDO101 and pHC101 probes, respectively, connected to a Hach HQ40d meter (Hach Company, Loveland, Colorado, USA). Temperature was measured with a digital thermometer. We placed HOBO® pH and temperature loggers (MX2501, Onset Computer Corporation, Bourne, Maryland, USA) in select test tanks and in the effluent tank for continuous measurement of temperature and pH. Total ammonia nitrogen was measured once a week in each test tank by spectrophotometric methods (Hach DR 3900, Hach, Loveland, Colorado, USA); un-ionized ammonia concentration was calculated using total ammonia nitrogen (NH 4 -N mg/L), temperature, and pH of each tank (Emerson et al. 1975). Alkalinity and hardness of the incoming water to the headbox were measured in triplicate at least once a week. Total alkalinity (mg/L as CaCO 3 ) was determined by spectrophotometric method (Hach DR 3900, Hach, Loveland, CO, USA) or by titrimetric method to a pH endpoint of 4.5 (APHA 2018b). Total hardness (mg/L as CaCO 3 ) was determined by spectrophotometric method (Hach DR 3900, Hach, Loveland, Colorado, USA) or EDTA titrimetric method (APHA 2018c).

Statistical analysis
We used the Statistical Analysis Software package (SAS, Ver 9.4, Cary, North Carolina, USA) for all statistical analyses, except macroinvertebrate and periphyton community comparisons (see below). We report P values for statistical tests that were conducted. The experimental unit was the tank. Means, ranges, standard deviation (SD), and 95% confidence intervals (CIs) were calculated for water quality measurements (DO, temperature, alkalinity, hardness, ammonia, specific conductance), and biomass on settlement plates. Data were checked for normality and homogeneity of variance (Shapiro-Wilk's and F-max; Sokal and Rohlf 1981). Periphyton biovolume was log transformed before model fitting. Periphyton cell count, biovolume, and native mussel growth (daily rate) were analyzed with a general linear model (Proc GLM) for main effects of CO 2 treatment. Least square means with Bonferroni adjustment identified differences among treatments. Native mussel condition indices were analyzed using a mixed model (Proc mixed). The model included the fixed effect of CO 2 treatment and tank nested within treatment as a random effect. Biomass (excluding zebra mussels) on settlement plates was analyzed using a repeated measures mixed model with simple covariance structure. Because samples were only collected from two of the four replicate tanks per treatment each week, the data from the first sampling periods for each tank comprised time period one (T1), and from the second sampling period for each tank comprised time period two (T2). Time period three (T3), which occurred on the final day of the experiment, comprised samples from all treatments and test tanks. The model included fixed effects of CO 2 treatment, sampling period, and the interaction of treatment sampling period with repeated measure on the tank. In all mixed models, differences among treatments were compared using least-squares means comparison with Bonferroni adjustment for multiple comparison.
Response of the macroinvertebrate and periphyton community to the CO 2 treatments was analyzed with Permutation Multivariate Analysis of Variance (PERMANOVA+ Prime-E version 7 software, Anderson et al. 2008). Twenty taxa groups were identified in the benthic biofilm on the settlement samples, representing three time periods during the CO 2 exposures. Data from the first sampling periods for each tank comprised time period one (T1), and from the second sampling period for each tank comprised time period two (T2). Time period three (T3), which occurred on the final day of the experiment, comprised samples from all treatment/replicate test tanks. The macroinvertebrate community count data were squareroot-transformed before applying Bray-Curtis similarity measures to determine the effects of treatment and time on communities within the tanks. Zebra mussels were found only in the C amb tanks and therefore, were not included in the community analysis. We conducted a repeatedmeasures mixed model PERMANOVA+ analysis of the Bray-Curtis similarities measure for main effects of CO 2 treatment and sampling period as fixed effects and tank nested within treatments as a random effect. Pairwise comparisons of treatment levels or sampling periods used a Monte-Carlo test with permutation of residuals under a reduced model. We report exact Monte-Carlo P-values unadjusted for number of comparisons. Similarity percentages (SIMPER) analysis was conducted to determine taxonomic group(s) driving the dissimilarities observed among treatments and/or time periods. Periphyton community analysis followed that of the macroinvertebrate analysis with the exception that the taxonomic resolution (48 groups) in the data was reduced to six divisions (Cyanophyta, Bacillariophyta, Chlorophyta, Haptophyta, Euglenophyta, and Cryptophyta) to reduce the number of unique taxa (1 observation per taxa) in the data matrix.

Zebra mussels
Veligers were most abundant in the incoming water during the first week of July (mean test tank density = 13.4 veligers/L, SD 3.9). Veliger density Figure 1. Mean dry biomass density of all macroinvertebrates (solid bars) and zebra mussels (dashed bars) on settlement plates in ambient, low, and high carbon dioxide (CO 2 ) tanks. No zebra mussels were collected from low and high CO 2 tanks. Settlement samplers were deployed on 06 June 2019 and four plates per tank were removed weekly from two tanks per treatment, except on 19 Aug 2019 when all tanks were sampled. Ambient = 2478 µatm PCO 2 ; Low = 35,094 µatm PCO 2 ; High = 66,685 µatm PCO 2 ; n = 4 tanks per treatment.
We found zebra mussels on settlement plates in C amb tanks starting on July 15 and each week thereafter until the end of the study ( Figure S4). We found no settlement on plates in CO 2 treatments except for a single zebra mussel in a C high tank. Cumulative mean (SD) settlement in C amb tanks ranged from 28.9 (3.9) to 67.1 (7.1) mussels per plate (0.10 to 0.23 mussels /cm 2 ) ( Figure S4). At the end of the experiment, the total zebra mussel count in whole tank samples (minus settlement plates) ranged from 552 to 1239 in C amb tanks, 0 to 4 (6 total) in C low tanks, and zero in C high tanks. When the infusion manifolds were dismantled at the conclusion of the study, the C amb manifold had zebra mussels encrusting the inside while the elevated CO 2 manifolds had no attached zebra mussels.
Total biomass on settlement plates in C amb tanks increased concomitant with the number and size of zebra mussels (Figure 1). Total zebra mussel biomass ranged from mean (SD) of 15.5% (3.3%) at the first sampling period to 94.9% (2.0%) at the last sampling period in C amb tanks. No zebra mussel biomass was found in C low and C high tanks. Overall, biomass (excluding zebra mussels) differed by treatment (degrees of freedom, df 2,9 F = 13.19, P = 0.002) and sample period (df 4,9 F = 7.75, P = 0.004); a treatment × sample period interaction was detected (df 4,9 F = 4.46, P = 0.011). Across sample periods, non-zebra mussel biomass was greater in C amb and C low tanks compared to C high tanks (C low vs C high t = 5.812; adj P = 0.002; Figure 1) and did not differ between C amb and C low tanks t = 2.22; adj P = 0.160). Non-zebra mussel biomass decreased at the end of the experiment compared to the midpoint sample period (t = 3.93, adj P = 0.003).

Native mussels
Native mussel survival was 100% in all tanks except for mortality of a single L. siliquoidea in a C high tank at the end of the study. At the end of experiment, L. cardium shell length (SD) averaged 34.2 (2.2), 34.3 (3.1), and 27.4 mm (4.5) in C amb , C low , and C high , respectively. Daily mean growth in shell length of both Lampsilis species was reduced by CO 2 exposure (F 2,9 = 29.21, P = 0.001) especially in C high (Figure 2A, Table S2). Daily mean shell growth of  ) and (B) high CO 2 treatment tank (mean 66,685 µatm PCO 2 ) at the end of the study. Note dull pigmentation of the periostracum (P) and erosion on the beak sculpture (*) in mussels exposed to CO 2 . New shell growth started at dark line (solid arrow in A) and band (dashed arrow in B). The wide band on shells in high CO 2 (B) indicates interrupted or reduced growth. Photographs by D. Waller.
Qualitatively, we saw loss of shell coloration and integrity in C high compared to C amb and C low treatments. Shells in C amb and C low tanks had saturated pigmentation of the periostracum and shiny nacre, while those in C high tanks appeared bleached and dull (Figure 3). In C high tanks, we saw loss of periostracum from the umbo, a wide growth band (rather than a line), and increased shell brittleness. The relative loss of shell mass was quantified with the calculated shell condition index (SC). We saw a treatment effect on SC of both L. cardium (df 2,9 F = 22.56, P < 0.003) and L. siliquoidea (df 2,9 F = 22.72, P < 0.003; Figure 2B, Table S2). Shell condition of L. cardium was reduced in C high compared to C amb (t = 6.14, adj P < 0.001), but not in C low (t = 0.71, adj P = 1.000). We detected differences in SC of L. siliquoidea between all treatments which was greatest between C amb and C high (t = 6.63, adj P < 0.001), followed by C low and C high (t = 4.40, adj P = 0.005). Shell condition in L. siliquoidea was similar in C amb and C low treatments (t = 2.23, P = 0.159) ( Figure 2B, Table S2). We detected a treatment effect on tissue condition (TC) of L. siliquoidea (df 2,9 F = 11.82, P = 0.003). Lampsilis siliquoidea TC differed between C amb and C high (t = 4.04, adj P = 0.009) and between C low and C high (t = 4.37, adj P = 0.005), but not between C amb and C low (t = −0.34, adj P = 1.000; Table S2). Likewise, we found a treatment effect on TC of L. cardium (df 2,9 F = 60.91, P < 0.001) but the analysis was limited to two mussels per tank. We detected differences in TC of L. cardium between C amb and C high (t = 9.56, adj P < 0.001) and C low and C high (t = 9.56, adj P < 0.001), but no difference between C amb and C low (t = 0.00, adj P = 1.000; Table S2).
Benthic community composition was assessed based on Bray-Curtis similarity measure and differed among the CO 2 treatment levels (Permanova-CO 2 , df 2,35 Pseudo-F = 12.933; P(perm) = 0.001), and among sampling periods (Permanova-Period, df 2,35 Pseudo-F = 6.091; P(perm) = 0.001) but did not indicate a treatment × time interaction (Permanova-CO 2 × Period, df 4,35 Pseudo-F = 0.718; P(perm) = 0.774). Because pairwise tests between treatment levels indicated the number of unique permutations was < 100 (unique permutations n = 35), pairwise Monte-Carlo (MC) tests (999 permutations) were used which indicated differences between the C high and C low treatments (t = 4.96, P(MC) = 0.001), C high and C amb (t = 3.51, P(MC) = 0.001), and C low and C amb (t = 1.89, P(MC) = 0.023). Similarity between C amb and C low was 68.04%, C low and C high was 53.57%, and C amb and C high was 59.3%. Similarity within CO 2 treatments was above 69.5% showing good agreement among replicate test tanks. Taxonomic groups consistently driving differences among treatments were Oligochaeta and Chironomidae with 13 to 26% of dissimilarities; Oligochaeta were commonly found in greater abundance within the C high treatments and in lesser abundance in C amb and C low treatments. Chironomidae were most common in C low treatments followed by C amb and C high treatments in order of abundance ( Figure 4). Cladocera was at 19.9% dissimilarities for C low and C amb treatments; however, dissimilarities to standard deviation ratio was less than 0.9 indicating high variability relative to contribution to the dissimilarities. Hydrozoa and Ostracoda each contributed 11 to 12% of the dissimilarities among treatments. Differences in the macroinvertebrate community between sampling periods used pairwise comparisons which indicated differences between T1 and T2 (t = 2.48, P(perm) = 0.004), T1 and T3 (t = 3.10, P(perm) = 0.006), and T2 and T3 (t = 2.15, P(perm) = 0.017). Similarity in macroinvertebrate community composition between sampling periods was as follows: 58.7% T1 vs T2, 67.3% T2 vs T3, and 59.7% T1 vs T3. Similarity within sampling periods was 60.9, 65.8, and 72.4% for T1, T2 and T3, respectively ( Figure S5).
Periphyton cell density and biovolume were influenced by CO 2 (cell density F 2,9 = 21.40, P < 0.001; logbiovolume F 2,9 = 6.49, P = 0.018) with greatest values occurring in C high , followed by C low and C amb treatments (Table 1, Figures S6 and S7). Periphyton measures did not differ between C low and C high treatments, but both differed from C amb . Forty-eight taxonomic groups were identified in the periphyton collected in our study. Because of the substantial number of unique taxa in one or a few samples, the data were aggregated at the taxonomic level of division. Six individual divisions were represented in this aggregation with Cyanophyta (blue-green algae), Bacillariophyta (diatoms), and Chlorophyta (green algae) most observed in the samples (Table 1, Figures S6 and S7). Haptophyta, Euglenophyta (euglenids), and Cryptophyta (cryptomonads) were also represented in the samples but not consistently across treatments (Table 1, Figure S6; note this analysis is based on counts per cm 2 ). Bray-Curtis similarity differed among the CO 2 treatment levels (Permanova-CO 2 , df 2,11 Pseudo-F = 5.55; P(perm) = 0.012). Pairwise differences were observed between C amb and C high (t = 2.61, P(perm) = 0.007), C amb and C low (t = 1.57, P(perm) = 0.008), but not between C high and C low (t = 1.09, P(per) = 0.347). Similarity within treatments was 51.6, 83.3, and 71.3%, for C amb , C low , and C high , respectively, indicating the greatest variability in C amb . Euglena sp. was the only taxon that was unique to C amb tanks, and Bacillariophyta and Chlorophyta comprised the greatest biovolume (Table 1). Cyanobacteria were absent from C amb tanks except a single species (Eucapsis sp.) in one C amb tank. Similarity between treatments was: 46.2% C amb vs C low, 74.3% C low vs C high , and 36.8% C amb vs C high . More than 70% of the cumulative dissimilarities between treatments were from three divisions, Cyanobacteria, Bacillariophyta, and Chlorophyta. In general, these divisions had the greatest mean abundance in C high treatment, followed by C low , and C amb treatments (Table 1, Figure S6). The abundance of all divisions increased in elevated CO 2, accompanied by a shift in community composition that was primarily due to Cyanobacteria, Bacillariophyta, and Chlorophyta (Table 1, Figure S6). Cyanophyta were more consistently found in the CO 2 treatments compared to the controls (C amb ).

Carbon dioxide and water quality
Carbon dioxide levels were measured by three methods and are reported as PCO 2 (µatm) and CO 2 concentration (mg/L) for comparability to other reported CO 2 studies ( Figure 5, Table S4); hereafter, we refer only to calculated PCO 2 because these values were derived from the most sampling points. PCO 2 levels ranged from 2,419 (SD 1,842) to 2,504 (SD 1,886) µatm in C amb tanks; 33,739 (SD 9,254) to 36,341 (SD 8,925) µatm CO 2 in C low tanks; and 65,289 (SD 14,761) to 68,715 (SD 16,069) µatm CO 2 in the C high tanks ( Figure 5, Table S4). Carbon dioxide levels fluctuated during the study but remained similar among replicate tanks within a treatment. The regression of logPCO 2 and pH showed strong correlation between the two parameters (logPCO 2 = 11.4 -1.003*pH, R 2 = 0.985). The effluent tanks in the test system were aerated to off-gas CO 2 and the PCO 2 of discharged water was ≤ ambient PCO 2 throughout the study. Mean alkalinity and hardness of the source water was 198.1 (SD 54.7; range 134-286) and 276.5 (SD 62.3; range 212-379) mg/L as CaCO 3 , respectively. A sharp decrease occurred from June 13 to 17, immediately Figure 5. Mean partial pressure (solid line) and concentration (dashed line) of carbon dioxide (CO 2 ) in ambient, low, and high treatments during exposure period. Error bars indicate standard deviation. Concentration was measured by titration with NaOH; PCO 2 was calculated from pH, temperature, and alkalinity (Robbins et al. 2010). n = 4 tanks per treatment.
after opening of the harbor gate to the main channel of the Mississippi River on June 13. Water from the main channel of the river circulated throughout the harbor after this date. Mean (SD) alkalinity and hardness after June 13 were 170.3 (22.8) and 241.5 (21.3) mg/L as CaCO 3 , respectively ( Figure S8).
Dissolved oxygen, temperature, and flow rate were similar among test tanks and treatments ( Figure S9A-C, Table S5). Mean DO was slightly lower in CO 2 treatment tanks compared to controls, but there was no detectable CO 2 treatment effect. On occasion, debris (e.g., snail) temporarily blocked the inflow line to a test tank, which reduced DO for brief periods. A power failure occurred in week 9 and flow was stopped for ~ 8 h; DO decreased to < 3.0 mg/L for an unknown portion of that time. There was a persistent decrease in DO concentrations in all test tanks in August that was unrelated to the power failure. Mean (SD) DO in C amb was 8.4 (1.3) mg/L in June and July and decreased to 6.3 (1.1) mg/L in August ( Figure S9A-C, Table S5). A similar decrease occurred in C low and C high treatments. Mean temperature increased only 0.8 °C during the same time. Mean (SD) pH in C amb tanks was 8.07 (0.26) compared to 6.84 (0.11) in C low and 6.56 (0.13) in C high tanks.
Ammonia is reported as total ammonia-nitrogen (TAN) and un-ionized ammonia (NH 3 -N). We normalized TAN to pH 8 (TAN Eq ) for comparison to water quality criteria. Ammonia concentrations were greater in control tanks compared to the CO 2 treatment tanks (Table S6); however, concentrations in all tanks were below chronic aquatic water quality criteria (USEPA 2013). At mean water temperature 25 °C and pH 8.1, the chronic criterion magnitude (CCC) is 0.49 mg TAN/L (USEPA 2013), about four-fold greater than mean concentrations in C amb tanks.

Zebra mussels
Carbon dioxide prevented settlement of zebra mussel veligers at ~ 35,000 µatm PCO 2 in our test system. Tanks that received CO 2 -infused water had zero or minimal zebra mussel settlement. Additionally, when the system was dismantled at the end of the study, we found that the two CO 2 infusion chambers were lined with a layer of organic biomass while the control manifold was completely encrusted with zebra mussels. In the whole tank contents, we found no live zebra mussels in C high tanks and a total of six live mussels in C low tanks. We did not determine whether CO 2 killed veligers at this level or only prevented settlement. Veligers flowed out of test tanks to effluent tanks at similar rates (Table S1); however, live zebra mussels were only found in effluent tanks receiving water from the C amb treatment, indicating that CO 2 was lethal and not just a temporary deterrence to veliger settlement. Even lower CO 2 levels may be effective on veliger settlement. Claudi et al. (2012) reported that veligers did not settle below pH 7.1 in water acidified with phosphoric acid. Based on our pH-CO 2 regression, we predict that settlement could be prevented near 20,000 µatm PCO 2 in our microcosm.
Mussel control programs that use chemicals such as chlorine and copper employ a range of treatment schedules from continuous to intermittent to single application depending on the targeted life stage and the objectives of the treatment (Mackie and Claudi 2010;Chakraborti et al. 2014). Here, we assessed continuous infusion of CO 2 and did not determine minimum exposure time to prevent settlement. Short-term exposure (96 h) of adult mussels to ~ 110,000 µatm PCO 2 at 20 °C reduced attachment by 75% . Likely, short, intermittent exposure to lesser CO 2 levels would be effective for preventing biofouling by the more sensitive veliger stage. Additional data on minimum lethal concentration and exposure period for veligers would be needed to develop alternative application schedules for CO 2 , like those for chlorine and copper compounds.

Native mussels
Chronic exposure to CO 2 was not lethal to juvenile native mussels at the levels that we tested. Other studies report varying degrees of mortality to juvenile mussels depending on the CO 2 level and exposure duration. For example, in ~ 4-d exposure to 100,000 to 120,000 µatm PCO 2 at 20 °C there was minimal mortality of juvenile L. cardium (5%) and Leptodea fragilis (7%). In contrast, Jeffrey et al. (2018b) reported ~ 30% mortality of juvenile L. siliquoidea after 14-d exposure to 58,000 µatm PCO 2 at 25 °C. Survival was dependent on size (shell length), and juveniles > 12.1 mm in length (range 7 to 16 mm) experienced no mortality. In 28-d exposure, the estimated lethal levels to 20% (LC20) of juvenile L. higginsii (Lea, 1857) and L. siliquoidea were 31,800 and 58,200 µatm PCO 2 , respectively (Waller et al. 2019) while we found 100% survival at these levels. We tested larger L. siliquoidea (mean shell length = 10.2 mm, range 5.6 to 14.9 mm) than were tested by Waller et al. (2019) (mean shell length = 7.2 mm, range 4.2 to 9.2 mm) which may account for some of the lethality differences between studies. Test systems and holding condition may also influence juvenile response to CO 2 . Earlier studies used laboratory water and fed juveniles an artificial diet (Jeffrey et al. 2018b;Waller et al. 2017Waller et al. , 2019, which may not be optimal for their growth and condition. In the present study, juveniles were held in river water and received a constant supply of a natural food source. The microcosm was more similar to the river environment, thus juveniles may have been in better condition and had greater tolerance to a stressor, like CO 2 , than those in a laboratory setting. Although CO 2 was not lethal to juvenile mussels in our study, sublethal effects on growth and condition were detected in the high treatment. The decline in tissue condition indicated that CO 2 adversely affected tissue growth. Energy (caloric) demands of juvenile mussels are especially high to support tissue growth. Increased metabolic demands of acid-base and ion regulation (Hannan et al. 2016a;Jeffrey et al. 2018a, b) may have diverted energy resources for tissue growth and maintenance in L. siliquodea in C high . Like previous studies, we found that CO 2 reduced mussel shell growth and mass, particularly in C high (66,685 µatm PCO 2 ). Waller et al. (2019) reported that shell growth was reduced at half this level (median effective concentration, EC50 = 28,6000 µatm PCO 2 ) in 28-d exposure of the same species. As previously discussed, differences among studies may be due to use of smaller animals (~ 4 mm) by Waller et al. (2019) compared to the present study. The shell is the source of carbonate for acid-base and ionic balance in mussels and a demonstrated consequence of elevated CO 2 in freshwater mussels is reduced biomineralization and shell formation Waller et al. 2019). The shell is the primary protection for mussels and reduced shell mass and integrity may increase their susceptibility to injury and predation.
Overall, the reported data on effects of CO 2 to native mussels show that low levels (~ 20,000 to 35,000 µatm PCO 2 ) are not lethal to juvenile mussels and would prevent zebra mussel settlement. However, if CO 2 infusion occurs continuously for several weeks, sublethal effects on juvenile mussel growth and condition are likely, even at this low range. Preferably, short and/or intermittent treatment levels can be developed that could reduce biofouling without adversely affecting juvenile mussels. Until more data are developed on size-related toxicity of CO 2 to native mussels, CO 2 use could be considered only for larger juveniles (> 12 mm shell length) and adult mussels. Furthermore, the existing data on CO 2 toxicity to juvenile native mussels is limited to lampsiline species. Additional CO 2 toxicity data are needed, especially for thin-shelled (i.e., Tribe Anodontini) species, to determine its safety more broadly to unionid mussels.

Benthic community
Overall benthic invertebrate biomass was similar between C amb and C low tanks and trended lower in C high tanks (Figure 1). From late July to early August, biomass (excluding zebra mussels) declined in C amb tanks as zebra mussel density increased. At the same time, biomass increased in C low tanks in the absence of zebra mussel competition. However, this same increase did not occur in C high tanks, despite the absence of zebra mussels, indicating an effect of CO 2 on benthic colonization at the higher level. Biomass was similar among all treatments in late August, despite a sharp increase in zebra mussel density in C amb tanks, indicating that macroinvertebrate reproduction and settlement likely peaked in early August. For example, we observed episodes of chironomid emergence in August that likely contributed to decreased biomass on the last sampling date.
Acidification from strong acids is widely reported to cause reductions in macroinvertebrate abundance and richness (e.g., Zischke et al. 1983;Allard and Moreau 1987;Feldman and Connor 1992). Together with reduced biomass, we expected a similar decrease in taxa richness in CO 2 treatments. We found that most riverine taxa in the microcosms tolerated extended exposure to hypercapnia. Only three taxa were unique to the C amb tanks, including two water mite genera, each with a single occurrence. Although reported pH tolerances of macroinvertebrate taxa vary widely, most are below the levels produced by CO 2 infusion in the microcosms. Moreover, the mean pH in C high (6.52 to 6.58) was above the minimum EPA water quality criteria of pH 6.5 (USEPA 2020). We might have detected shifts in species composition within groups with finer taxonomic resolution. More sensitive bioindicator taxa, such as Ephemeroptera, Plecoptera, Trichoptera, were represented by only two genera in the microcosm (Figure 4, Figure S5). Potamyia (Trichoptera) and Caenis (Ephemeroptera) were found in both C amb and CO 2 tanks, but in low abundance, so we were unable to detect a response of these taxa to CO 2 treatments.
Differences in macroinvertebrate community composition among treatments were driven primarily by Chironomidae, Oligochaete, Cladocera, Ostracoda, and Gastropoda (i.e., Gyraulus, Planorbidae, Physella). These groups of invertebrates, particularly crustaceans, oligochaetes, and chironomids, can comprise a major portion of consumer biomass in aquatic habitats and are major food sources for larger invertebrates and vertebrates. Significant shifts or reduction in their abundance that might occur from CO 2 infusion could adversely affect higher trophic levels and nutrient cycles. The two dominant macroinvertebrate taxa, Chironomidae and Oligochaetes, are adapted to a wide range of environmental conditions including hypoxia and organic pollution (Beck 1977;Lindegaard 1995;Rodríguez and Reynoldson 2011). Both taxa have respiratory pigments that provide buffering capacity in low pH water and resistance to the hypoxia which occur with elevated CO 2 . Therefore, the relatively high abundance of these taxa in the CO 2 treatments was not surprising. Chironomids were less abundant than oligochaetes in C high treatment which indicates greater tolerance of the latter to chronic hypercapnia.
The effects of CO 2 on primary producers in freshwater has focused on phytoplankton, and generally demonstrated increased growth in response to elevated CO 2 (Low-Décarie et al. 2011, 2014Verschoor et al. 2013;Shi et al. 2015Shi et al. , 2017.We measured chlorophyll a to monitor phytoplankton concentration but we did not detect a change because of the short retention time in the test tanks. We also assessed the effects of CO 2 on the periphyton component. Periphyton cell density was greater in CO 2 treatments than the ambient but was not different between CO 2 treatments. The abundance of all divisions increased in CO 2 , but there was a shift in composition from predominately diatoms and chlorophytes in ambient tanks to a greater abundance of cyanobacteria in CO 2 treatment tanks. Cyanobacteria were found in only one C amb tank but were relatively abundant in all CO 2 treatment tanks ( Figures S6 and S7). A potential increase in cyanobacteria abundance as a result of CO 2 infusion could adversely affect aquatic invertebrates by reducing food quality (Müller-Navarra et al. 2000;Basen et al. 2012) and producing harmful toxins . In general, nutrients and light being the same, additional CO 2 at levels used in this study would increase periphyton abundance and may favor less desirable groups such as cyanobacteria (Visser et al. 2016), while short-term treatments of CO 2 would likely result in only temporary increase in these groups.

Conclusions
We demonstrated that CO 2 is an effective tool for preventing zebra mussel settlement at ~ 35,000 µatm PCO 2 . Further investigation is needed to establish minimum effective CO 2 concentrations and exposure duration to develop alternative application schedules that would maximize effectiveness, reduce costs, and minimize nontarget effects. Infusion systems for CO 2 in closed systems could be readily adapted from ours or a similar design. In open water, A CO 2 curtain could be used to reduce settlement/biofouling on water infrastructure (e.g., sumps, trash racks), around marinas (e.g., docks) and in isolated ponds or bays. In closed or open water application, CO 2 can be off-gassed by aeration. Before implementing CO 2 to prevent biofouling around native mussels, more toxicity data are needed for smaller unionid juveniles and additional species. Efficacious concentrations of CO 2 were not lethal to juvenile native mussels, but chronic exposure to the levels tested here could reduce juvenile growth and long-term viability. Our study conducted during peak reproductive period for many freshwater taxa was unique in assessing the cumulative effects of chronic hypercapnia on the benthic community. Elevated PCO 2 did not eliminate any major benthic groups but shifted abundance to more tolerant taxa. Periphyton abundance and richness increased across divisions, most notably in cyanobacteria, diatoms, and chlorophytes. As CO 2 is implemented in AIS control programs, more extensive investigations would allow managers and regulatory professionals to better understand the primary response of populations and secondary effects on aquatic communities from elevated CO 2 to avoid unintended adverse consequences of its use.

Supplementary material
The following supplementary material is available for this article: Figure S1. Schematic of test system. Figure S2. Carbon dioxide infusion chamber and test tanks. Figure S3. Veliger density in inflow to test tanks. Figure S4. Cumulative zebra mussel settlement in ambient carbon dioxide tanks. Figure S5. Boxplots of six most abundant macroinvertebrate taxa in ambient, low and high carbon dioxide treatment at three sampling periods. Figure S6. Boxplots of benthic algae abundance by division in test tanks. Figure S7. Heatmap of benthic algae abundance by division in ambient, low and high carbon dioxide treatment. Figure S8. Mean alkalinity and hardness of source water during carbon dioxide exposure period. Figure S9. Daily mean measurements of (A) dissolved oxygen, (B) temperature, and (C) water flow rate in ambient, low, and high carbon dioxide treatment during the exposure period. Table S1. Estimated mean, standard deviation, range, and the percent of retained zebra mussel veligers in test tanks. Table S2. Mean (standard deviation) shell growth and condition indices of juvenile mussels in ambient and carbon dioxide treatments. Table S3. Macroinvertebrate taxa total count and percent occurrence. Table S4. Mean, standard deviation, and range of carbon dioxide concentration and partial pressure in treatments during the study period. Table S5. Mean, standard deviation, and range of dissolved oxygen, pH, temperature, and flow rates in ambient and treatment tanks. Table S6. Mean, standard deviation, and range of total ammonia-Nitrogen, un-ionized ammonia and TAN equivalent during study period.